《1993 A lift-and-project cutting plane algorithm for mixed 0–1 programs》.pdf

Mathematical Programming 58 (1993) 295-324 295 North-Holland A lift-and-project cutting plane algorithm for mixed 0-1 programs Egon Balas, Sebasti~in Ceria and G6rard Cornu6jols Graduate School of Industrial Administration, Carnegie Mellon University, Pittsburgh, PA, USA Received 6 November 1991 Revised manuscript received 26 May 1992 We propose a cutting plane algorithm for mixed 0-1 programs based on a family of polyhedra which strengthen the usual LP relaxation. We show how to generate a facet of a polyhedron in this family which is most violated by the current fractional point. This cut is found through the solution of a linear program that has about twice the size of the usual LP relaxation. A lifting step is used to reduce the size of the LPs needed to generate the cuts. An additional strengthening step suggestedby Balas and Jeroslow is then applied. We report our computational experience with a preliminary version of the algorithm. This approach is related to the work of Balas on disjunctive programming, the matrix cone relaxations of Lov~sz and Schrijver and the hierarchy of relaxations of Sherali and Adams. Key words: Cutting planes, projection, mixed 0-1 programming, disjunctive programming. I. Introduction More than thirty years have elapsed since the emergence of cutting plane algorithms for mixed integer programming, but branch and b o u n d is still the prevalent way to handle such problems. In the last 10-15 years there has been considerable progress in using combinatorial cutting planes for certain classes of pure integer programming problems, such as the symmetric traveling salesman problem, in c ombina t i o n with branch and b o u n d (see, for instance [18]). The success of this approach, known as